On Thu, 30 Oct 2014, Joakim Hernberg wrote:
On Wed, 22 Oct 2014 14:30:21 -0700 (PDT)
Len Ovens <len@xxxxxxxxxxxxx> wrote:
I have not done that much system tuning either. I did choose the i5
on purpose for no HT, chose to use a pci AI and which slot it should
be in. No wifi, performance mode. Mouse and kb are not USB (in fact
no USB use at all, now that I think about it) Video is Intel based.
I mainly use an i7, I did however disable HT :) I did not see many
xruns caused by it in my testing, but they were occasionally there.
The cost was noticably longer build times for wine and the kernel :)
I am not sure exactly how HT works. That is I do not know the mechanism
for switching threads. I would assume though, that it is cycle based,
after so many cycles we switch to another thread. That being the case, a
faster processor would allow lower latency with HT on. My experience with
a 2.4Ghz P4 was that jack at 64/2 was not really a problem with HT on but
anything lower was. I would expect an i7 running at 3.6Ghz would at least
be able to run jack at 32/2 with no problems, but at some point HT will be
a barrier to lower latency. In audio, it would seem that so long as there
are no xruns at the latency being used there is not problem. So long as
the sound "card" is providing the media clock that is true. If someone was
trying to run a stable PTP based sw media clock, I think jitter would be
a problem. I think that is one of the reasons most AoIP setups suggest
using their NIC/AI which would have it's own HW based PTP server. (a
number of NIC chips have one built in, even cheap $5 ones)
So far as I know, HT thread switching is controled by firmware outside the
control of the OS. If it could be controlled by the OS, low latency might
be enhanced with HT on as a high priority thread could be left in CPU
memory while servicing a lower priority thread. I don't think a single
core would be helped with this, but a 4 core cpu might. It would depend on
the total number of threads and how many of them had raised priority.
--
Len Ovens
www.ovenwerks.net
_______________________________________________
Linux-audio-user mailing list
Linux-audio-user@xxxxxxxxxxxxxxxxxxxx
http://lists.linuxaudio.org/listinfo/linux-audio-user
